Content
82%Weight 40%Scale 1-5Reviews the quality of instructions and guidance provided to agents. Good implementation is clear, handles edge cases, and produces reliable results.
The body is a compact, highly actionable runbook with executable commands for each subcommand and sensible validation/error exits. It stays lean and well-structured, with only minor conciseness and verification gaps.
Suggestions
Drop the opening Japanese sentence that restates the description to save tokens and avoid duplication.
After --user-default / --project-default mutations, add an explicit verify step (beyond --show) that confirms the resolved backend actually changed, e.g. re-running resolve-impl-backend.sh --role worker.
Tighten the long inline comment block on the cursor-agent binary discovery to a one-line rationale.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is mostly lean executable shell code and assumes Claude's competence without explaining Cursor or backend concepts, but the opening Japanese sentence partly restates the description and a few inline comments could be trimmed; efficient rather than maximally lean. | 4 / 5 |
Actionability | Provides fully executable, copy-paste-ready bash for every subcommand (--check, --user-default, --project-default, --unset) referencing concrete helper scripts, covering the common cases completely. | 5 / 5 |
Workflow Clarity | Clear sequence with explicit validation: HARNESS_PLUGIN_ROOT resolution errors out (exit 2), --check validates the cursor-agent binary (exit 3), mutations are followed by --show, and --unset confirms scope; minor gaps remain in re-verifying that a mutation actually took effect. | 4 / 5 |
Progressive Disclosure | Well-organized into Quick Reference, Rules, Flow, and Output sections with no nested or buried references; the skill is self-contained with no bundle files, so structure is good with only minor organization gaps in the long inline code blocks. | 4 / 5 |
Total | 17 / 20 Passed |